Theorem Proving on CombiBench
8Proof LengthClaude 4.6 Opus
Evaluation Results
| Method | Links | |
|---|---|---|
| Claude 4.6 OpusType=Proprietary LLM2026.04 | 8 | |
| Gemini 3.1 ProType=Proprietary LLM2026.04 | 15.8 | |
| GPT-5.3-CodexType=Proprietary LLM2026.04 | 22.3 | |
| Gemini 2.5 ProType=Proprietary LLM2026.04 | 23.8 | |
| DreamProverBackbone=Gemini 2.5 Pro, Type=Lemma Learning System2026.04 | 29.6 | |
| DreamProverBackbone=Gemini 3.1 Pro, Type=Lemma Learning System2026.04 | 31.3 | |
| DreamProverBackbone=GPT-5.3-Codex, Type=Lemma Learning System2026.04 | 37.4 | |
| DeepSeek-Prover-V2-7BType=Open-source LLM2026.04 | 41.4 | |
| Goedel-Prover-V2-8BType=Open-source LLM2026.04 | 45.3 | |
| HilbertBackbone=Gemini 3.1 Pro, Type=Agentic System2026.04 | 63.4 | |
| HilbertBackbone=GPT-5.3-Codex, Type=Agentic System2026.04 | 76.5 | |
| Goedel-Prover-V2-32BType=Open-source LLM2026.04 | 76.7 | |
| HilbertBackbone=Gemini 2.5 Pro, Type=Agentic System2026.04 | 85.7 |